From intelligence to autopoiesis: rethinking artificial intelligence through systems theory

The rapid advancements in the field of artificial intelligence (AI) have reinvigorated profound debates on the nature of intelligence, consciousness, and communication. Large language models (LLMs), in particular, are at the center of these discussions, as they generate complex linguistic patterns and challenge the traditional distinction between machine computation and human understanding. While LLMs are often seen as highly advanced statistical systems that generate text based on probabilistic patterns, both laypeople and experts tend to attribute human-like qualities to them.This article analyzes AI, particularly LLMs, from a systems-theoretical perspective and examines the extent to which these models can be understood as autopoietic, operationally closed systems. Building on Luhmann's system theory, it is argued that classical Turing machines are not sense-making systems, as they lack both self-reference in the sense of re-entry and the ability to make contingent selections from possibilities. In contrast, artificial neural networks (ANNs) exhibit a novel, loosely coupled interaction with social systems, as they can extract patterns from societal communication. This form of coupling differentiates them from classical software and positions them as hybrid systems that, while lacking their own mental states, are nonetheless deeply embedded in the structures of societal meaning production. The paper argues that LLMs should neither be regarded as purely technical tools nor as genuine cognitive entities. Instead, it proposes understanding their functioning as a new form of artificial meaning production—not as independent thinking, but as a recursive reflection of socially shaped linguistic patterns. This perspective not only opens new insights into the relationship between humans and machines but also calls for a critical reflection on how AI technologies are transforming our understanding of communication and cognition.